how does rpart handle missing values in predictors?

Question

From the ?rpart documentation -

na.action : the default action deletes all observations for which y is missing, but keeps those in which one or more predictors are missing.

How does it impute missing values in predictors?

score 5 · Answer 1 · edited Jul 18 '14 at 18:19

5

This is where the surrogate variables come in - for each split, observations where the split variable is missing are split based on the best surrogate variable, if that's missing by the next best and so on, this is detailed in:

Therneau, Terry M. & Atkinson Elizabeth J. (March 28, 2014). An Introduction to Recursive Partitioning Using the RPART Routines, Mayo Foundation, section 5.

The document is accessible through rpart help (pdf).

edited Jul 18 '14 at 18:19

gung - Reinstate Monica

132,789
81
357
650

answered Jul 18 '14 at 18:00

milan

51
1
2

2

I am struggling to understand this sentence: "or each split, observations where the split variable is missing are split based on the best surrogate variable". Observations where the split variable is missing are split? How is it possible to split a variable is missing? Also I am not sure about the definition of a split variable. – par Jun 29 '16 at 12:57
This by no means answers your question completely, but maybe my answer in a related question illustrates the idea better. The linked documentation in the above answer explains in detail how the surrogate split is calculated. https://stats.stackexchange.com/a/345033/ – Alex May 08 '18 at 05:16

how does rpart handle missing values in predictors?

1 Answers1

Linked